Cognitive Flexibility

Abstract

Cognitive flexibility is a type of cognition: the capacity to adjust thinking and behaviour when the demands of a task or of the environment change. Laboratory research indexes it mainly through task switching and card sorting, and questionnaires add a self-report view whose agreement with those tasks is weak. Changing tasks slows responding, and advance preparation shrinks this switch cost without abolishing it. Flexibility is usually counted among the core executive functions; it matures slowly through childhood, changes in later life, depends on distributed frontal, parietal and subcortical networks, and is impaired across several clinical conditions. Three interactive demonstrations let readers run a cued task-switching simulation, sort cards under a hidden rule that changes, and compute mixing and switch costs from reaction times.

Keywords: cognitive flexibility, task switching, switch cost, set shifting, executive functions

Everyday life rarely holds one goal for long. A driver who meets a closed road, a student who moves from algebra to an essay and a clinician who must drop a first diagnosis all need to release one way of working and take up another. The research literature calls this capacity cognitive flexibility, yet it does not define the term in a single way: a review of the neuroimaging evidence called for more consistent operationalization of it in clinical and developmental neuroscience (Dajani & Uddin, 2015), and Ionescu (2012) devoted a paper to examining what kind of capacity flexibility is.

Most reviews place it among the executive functions, the effortful control processes that guide behaviour toward goals. In Diamond's review, flexibility is one of three core executive functions, alongside working memory and inhibition (Diamond, 2013). This article covers how flexibility is measured, what the switch cost reveals, how the capacity develops and declines, its neural basis, its clinical relevance and the evidence on training it.

What Cognitive Flexibility Is

The ability to effectively adapt one's cognitive and behavioral strategies in response to changing task or environmental demands.

Scope note for the MeSH descriptor Cognitive Flexibility (D000098282), U.S. National Library of Medicine.

MeSH lists one entry term under this descriptor, Cognitive Inflexibility. It names the other side of the same capacity: difficulty in letting go of a rule, task or way of thinking once it has stopped serving, which this article takes up in its section on clinical relevance.

The construct has no single agreed operationalization. Dajani and Uddin, reviewing neuroimaging work on set shifting and task switching, argued that the construct needs to be operationalized more consistently in clinical and developmental research, and they left unresolved whether flexibility depends on a dedicated neural basis separate from the executive control network or instead emerges from interplay among nodes within that network and other networks (Dajani & Uddin, 2015). A later review treated cognitive and behavioural flexibility as closely intertwined and described the task impurity problem: no laboratory measure of flexibility draws on flexibility alone (Uddin, 2021).

Most researchers nonetheless treat flexibility as one of a small set of executive functions. Diamond (2013) counted it among three core functions together with working memory and inhibition, the last covering both the suppression of responses and the control of interference. That review flagged two unsettled questions, how executive functions relate to intelligence and how separable working memory and inhibition really are, and noted that stress, poor sleep and loneliness impair executive functions while practice can improve them. A later chapter by the same author added a measurement warning: a task seldom engages a single executive function, so failing it may reflect a weakness in some other function (Diamond, 2020).

On the unity-and-diversity account, executive functions are correlated but separable abilities. It was built on a latent-variable study of one hundred thirty-seven college students who completed tasks such as the Wisconsin Card Sorting Test and the Tower of Hanoi; shifting, updating and inhibition emerged as moderately correlated yet separable factors, a dual-tasking measure was unrelated to all three, and the sample was limited to college students (Miyake et al., 2000). A later review of individual-differences work kept the picture of functions that are correlated but distinct and added that executive functions are not the same thing as general intelligence (Friedman & Miyake, 2017). The account has been challenged, though: a systematic review that re-analysed 21 correlation matrices by bootstrap resampling found no single model that fitted every sample; adult samples showed both unity and diversity, with the nested-factor model accepted and selected most often, while child samples most often favoured a single factor, with some support for two-factor models lacking a separate shifting factor, and the authors warned that underpowered samples and possible publication bias may have produced models that do not replicate (Karr et al., 2018). Later work asked whether shifting itself is a single ability. Modelling 20 task pairs in 119 young adults, von Bastian and Druey (2017) found that five distinct but mostly intercorrelated shifting factors fit best, two of which could not be told apart statistically from a higher-order general shifting factor; stimulus-set shifting related only to mapping shifting, and the sample contained young adults only. Whether shifting is one ability or a family of related ones is therefore an open question.

A related idea from clinical psychology, psychological flexibility, is broader than the laboratory construct. A narrative review placed it at the centre of psychological health and linked it with acceptance, mindfulness and emotion regulation, while describing the research base as fragmented and calling for measurement in dynamic, real-world settings (Kashdan & Rottenberg, 2010).

Task Switching and the Switch Cost

Changing tasks costs time. Rogers and Monsell (1995) had participants classify digits as odd or even and letters as consonants or vowels, with the task changing predictably every second or every fourth trial, the alternating-runs paradigm. Responses were slower on trials that followed a change, an increase in reaction time known as the switch cost. A longer interval between the response and the next stimulus reduced the cost, but some of it was left even with 1.2 s to prepare, and this residual appeared only on the first trial of the new task.

Cued designs gave the same answer with tighter control over timing. Meiran (1996) presented a cue naming the task before each two-dimensional stimulus and varied the cue-target interval; the task-shift cost fell as preparation time grew, and the reduction could not be explained by the time that had passed since the previous trial. Rule structure matters as well. In four experiments using geometric classification and arithmetic tasks, switching-time costs rose with rule complexity and changed with task cuing, and the two factors combined additively, which pointed to separate stages of goal shifting and rule activation; switching from a familiar to an unfamiliar task also appeared to be slower (Rubinstein et al., 2001).

The task-set reconfiguration account explains these findings as the price of resetting the cognitive system for a new task, and its standing is contested. Monsell (2003) reviewed the evidence that preparation lowers the switch cost but never removes it, discussed task-set inhibition alongside reconfiguration, and judged the neuroimaging evidence unresolved because studies had not yet separated the regions that exert control from those being controlled. Before that, Allport, Styles and Hsieh had taken up the dynamic control of tasks in a chapter on shifting intentional set (Allport et al., 1994). As a later review recounts, Allport and colleagues are tied to a rival account, task-set inertia, which treats the switch cost as proactive interference carried over from mappings that were in force earlier (Kiesel et al., 2010). Kiesel and colleagues surveyed five basic paradigms, mainly from the behavioural literature, and, while describing hybrid accounts in which switch costs reflect both reconfiguration and interference, concluded that it was not yet clear whether executive control processes must be postulated to explain switch costs, at least in task cuing; they also noted that mixed-task blocks impose a heavier memory load than single-task blocks, which confounds comparisons between them (Kiesel et al., 2010). Vandierendonck and colleagues likewise traced the cost to reconfiguration plus interference control, agreed that preparation time lowers it, and found that interference explains only part of it (Vandierendonck et al., 2010). A further challenge targeted the cuing paradigm directly: separating repetitions of the cue from changes of task, Logan and Bundesen (2003) argued that cued switching performance can be explained without any endogenous act of control. An integrative review now places task switching inside a broader treatment of multitasking (Koch et al., 2018).

Another mechanism that contributes to the cost is backward inhibition. Across six experiments, Mayr and Keele (2000) found evidence that a task set just abandoned is held down by inhibition, which makes an early return to it harder, and this effect can account for at least part of the residual cost. Kiesel and colleagues reviewed the n-2 task-repetition cost as a more convincing index of this persisting inhibition than asymmetrical switch costs (Kiesel et al., 2010).

Switch costs persist even when participants choose the task themselves. In a voluntary task-switching study with single digits judged for parity or magnitude, Arrington and Logan (2004) observed a switch cost of 310 ms at a 100-ms response-to-stimulus interval and 94 ms at a 1,000-ms interval; participants also repeated tasks more often than random choice would produce.

Task switching has mostly been studied apart from dual-task performance, the domain that includes the psychological refractory period. Koch and colleagues brought the two research lines together in a review organised around cognitive structure, flexibility and plasticity (Koch et al., 2018).

The first demonstration simulates a cued switching experiment. Digits are judged for parity or magnitude according to a cue, and the preparation interval and run structure can be set before the run. A simulated observer's mean times on repeat and switch trials are plotted, so the switch cost can be seen to shrink with preparation toward a residual; the output is seeded and illustrative, not data.

Figure 1 shows the same pattern schematically: the gap between switch and repeat trials narrows as the preparation interval lengthens, but it does not close.

Set Shifting and Card Sorting

Card sorting is a long-established laboratory test of set shifting. Grant and Berg (1948) published a behavioural analysis of reinforcement and of how easily people shift to new responses in a Weigl-type card-sorting problem, the task that became the Wisconsin Card Sorting Test. Milner (1963) studied how different brain lesions affect card sorting and what part the frontal lobes play.

That frontal reputation has been brought into question. A critical update covering four decades of use concluded that the Wisconsin Card Sorting Test, in its traditional form, separates frontal from non-frontal damage poorly, that neuroimaging during the test shows widespread activity including non-frontal regions, and that an anatomically pure test of this kind cannot be built (Nyhus & Barceló, 2009). Imaging also shows that the test has internal structure. In an event-related fMRI study of eleven participants, Monchi and colleagues distinguished four stages of the task and found distinct circuits engaged at different stages: mid-dorsolateral prefrontal cortex (area 9/46), mid-ventrolateral prefrontal cortex (area 47/12), the caudate nucleus, the putamen and the mediodorsal thalamus, with negative feedback and working memory monitoring among the processes examined; the sample was small (Monchi et al., 2001).

Children are tested with a simpler sort. The Dimensional Change Card Sort asks a child to sort cards by one dimension and then, in a post-switch phase, by the other; continuing with the first rule is perseveration, and a border version extends the procedure (Zelazo, 2006). The Trail Making Test, another staple among neuropsychological tests, is often read as a shifting measure. In forty-one healthy older adults, performance on Part B was best predicted by working memory, with task switching next (Sánchez-Cubillo et al., 2009), a result from a small and healthy older sample.

The second demonstration is a feedback-driven card sort. Cards are sorted by colour, shape or number on feedback alone, and the hidden rule changes after a run of correct sorts. The display counts the trials needed to find the new rule and the perseverative errors, sorts that still follow the old rule after it has changed; it is illustrative and not a clinical test.

Questionnaire Measures

Self-report scales ask people how flexibly they think rather than timing them. The Cognitive Flexibility Scale was validated in two studies of 247 and 275 participants and was reported to be reliable and valid, with its validity examined against measures such as communication flexibility, self-monitoring and attitudinal rigidity (Martin & Rubin, 1995); the coefficients are not available in the abstract. The Cognitive Flexibility Inventory is a brief self-report instrument designed to measure seeing difficult situations as controllable, perceiving alternative explanations and generating alternative solutions, and its items formed a reliable two-factor structure; across two studies it showed reliability over 7 weeks and preliminary evidence of convergent validity, and it had not been tested in clinical groups (Dennis & Vander Wal, 2010).

Questionnaires and laboratory tasks do not measure the same thing in any straightforward sense. Laboratory and self-report indices of flexibility correlate only weakly (Uddin, 2021), so a high score on one does not license a claim about the other.

Development and Ageing

Switching matures late in child development. A cross-sectional study of 325 participants aged 4 to 13 years plus adults, roughly 30 per age, measured global, local and asymmetric switch costs on a computerized battery and found that switching was still immature at 13 (Davidson et al., 2006). The same data complicated the separation of executive functions, because reaction-time measures of memory and inhibition correlated, and the youngest children responded impulsively and lost accuracy.

Across the whole lifespan, switch costs trace a U shape. In 152 people aged 7 to 82, costs were larger in children and older adults than in young adults; age-related variance in switching was at least partly independent of age-related variance in perceptual speed and working memory, and practice reduced the costs, especially in children (Cepeda et al., 2001). A meta-analysis of 26 articles and 36 independent participant groups found a marked age deficit in global costs, the mixing cost of keeping two tasks ready, while local costs, the switch cost proper, grew multiplicatively with task complexity but showed no age deficit specific to attention; cuing and predictability did not alter the age effect, and only latency was analysed (Wasylyshyn et al., 2011).

Training effects also vary with age. Task-switching training in children aged 8–10, young adults aged 18–26 and older adults aged 62–76 produced near transfer and far transfer, including to fluid intelligence; near transfer was largest in children and older adults, and variable training impaired near transfer in children (Karbach & Kray, 2009). The wider brain-training evidence calls for caution here: a systematic review of the published studies that training companies cited found gains on the trained tasks but little far transfer (Simons et al., 2016). It also reported design flaws in many of those studies, found that none met every best practice, and judged the evidence for benefits to everyday performance weak (Simons et al., 2016).

Brain Basis

Animal work isolated a frontal contribution to shifting. Rats with ibotenic acid lesions of medial frontal cortex were impaired at extradimensional shifts, changing which stimulus dimension predicts reward, while acquisition and reversal learning were intact; the authors concluded that rat medial frontal cortex is functionally similar to primate lateral prefrontal cortex (Birrell & Brown, 2000).

In humans, a hybrid blocked and event-related fMRI design separated sustained from transient control during task switching, relating sustained right anterior prefrontal activity to the mixing cost and transient left superior parietal activity to the switching cost (Braver et al., 2003); the abstract does not give the sample size. Network studies widen the picture. Uddin (2021) linked flexibility to the lateral frontoparietal, midcingulo-insular and medial frontoparietal networks together with frontostriatal circuits, discussed variability in brain signals as a marker of flexibility, and noted that no interventional study had yet shown that changing brain network dynamics supports successful flexibility training. A chapter on executive functions similarly argued that they depend on regions beyond the frontal cortex (Cristofori et al., 2019), in line with the card-sorting evidence above.

Two proposals extend the biology. Anacker and Hen (2017) proposed that adult hippocampal neurogenesis supports cognitive flexibility, because inhibition of the dentate gyrus by adult-born neurons may reduce memory interference and enable reversal learning, and that such flexibility may in turn reduce anxiety-like and depressive-like behaviour. At a different timescale, a review of the dynamics of cognitive control drew on ERP, spectral and fMRI work to trace how control unfolds in time and listed open questions about how its dimensions fit together (Gratton et al., 2018).

Inflexibility and Clinical Relevance

Cognitive inflexibility, the MeSH entry term for this descriptor, is the clinical face of the construct: persistence with a rule or way of thinking after it has stopped working. A systematic review of 50 studies searched for links between rigidity and 18 transdiagnostic processes that maintain psychopathology; four of them (rumination, perfectionism, impulsivity and compulsivity) were reviewed systematically, and most of the included studies found rigidity correlated with, co-occurring with or predictive of them (Morris & Mansell, 2018). The authors read the pattern as consistent with inflexibility maintaining disorder, but they did not test that idea directly, and the processes were only inconsistently tied to distress.

Specific groups show the deficit on standard tasks. Performance on the Dimensional Change Card Sort is impaired in children with ADHD and in autism (Zelazo, 2006). Such deficits are not specific to one disorder: flexibility impairments occur in many kinds of psychopathology and may be transdiagnostic intermediate phenotypes (Uddin, 2021). Mood is a further point of contact. When the Cognitive Flexibility Inventory was developed, its validity was examined against the BDI-II, a self-report inventory of depression symptoms, though the inventory itself had not yet been tested in clinical populations (Dennis & Vander Wal, 2010). At the broadest level, psychological flexibility has been treated as central to psychological health (Kashdan & Rottenberg, 2010), a claim that rests on a narrative review of a fragmented literature.

Training and Everyday Flexibility

Two studies link flexibility to practice and performance outside the laboratory, and both are correlational. Experienced practitioners of mindfulness meditation outperformed meditation-naive controls on measures of cognitive flexibility and attention, including Stroop interference and the d2 concentration test, but the comparison was cross-sectional and the abstract does not report group sizes (Moore & Malinowski, 2009). Among expert decision makers tested in a laboratory, greater cognitive flexibility went with better decision making performance (Laureiro-Martínez & Brusoni, 2018).

A learning-based view has also been advanced. Braem and Egner (2018) proposed grounding flexibility in associative learning, so that contextual cues, including subliminal ones, and incentives could come to shape switching. On direct training, the evidence is mixed and transfer tends to be narrow, exercise effects are small, and there is no consensus on whether bilingualism confers an advantage (Uddin, 2021).

The third demonstration is a cost calculator. Mean reaction times for single-task blocks and for repeat and switch trials in mixed blocks can be entered or adjusted, and the display computes the mixing cost and the switch cost, starting from the values in the worked example below; the arithmetic is illustrative.

Discussion

Several findings are secure. Switching tasks slows performance, and preparation reduces the cost without removing it (Monsell, 2003); later reviews concur that preparation time lowers the cost (Vandierendonck et al., 2010). Flexibility changes across the lifespan, maturing slowly in childhood (Davidson et al., 2006) and showing an age-related deficit in mixing costs (Wasylyshyn et al., 2011), and in adult samples it can be separated statistically from other executive functions, although child samples more often fit a single factor (Karr et al., 2018).

The open questions are as substantial. What the residual cost reflects is unsettled: one review found ample evidence of task interference but judged it not yet clear whether executive control processes are needed to explain switch costs (Kiesel et al., 2010). Shifting itself may not be a single ability, since individual-differences modelling in young adults yields several related factors (von Bastian & Druey, 2017). Laboratory tasks and questionnaires correlate weakly, and training in one aspect of executive function generally yields short-term, narrow transfer, although one task-switching study reported far transfer (Uddin, 2021; Karbach & Kray, 2009), so it is unclear which measure an intervention should target. Underneath all of these lies the definitional problem with which the field began: studies still operationalize flexibility in different ways (Dajani & Uddin, 2015). Settling these questions would need studies that give the same people laboratory switching tasks and questionnaires, model both at the latent level and test transfer against active control conditions; this article does not claim that such studies have been run.

Current Directions

Recent work has moved in several directions at once. Network neuroscience treats flexibility as a property of interacting large-scale brain systems and asks how their dynamics vary (Uddin, 2021), while research on the timing of control examines how it unfolds over time (Gratton et al., 2018). Learning-based proposals ask how cues and incentives come to shape switching (Braem & Egner, 2018), and integrative reviews join task switching with dual-task research under the heading of multitasking (Koch et al., 2018).

Individual-differences research is refining the structure of executive functions, separating them from general intelligence (Friedman & Miyake, 2017) and breaking shifting into components (von Bastian & Druey, 2017). Reviews of executive functions stress their distributed neural basis (Cristofori et al., 2019) and the impurity of the tasks used to measure them (Diamond, 2020). Applied lines connect flexibility to clinical rigidity (Morris & Mansell, 2018), to a proposed role for hippocampal neurogenesis (Anacker & Hen, 2017) and to the decisions of experts (Laureiro-Martínez & Brusoni, 2018).

Common Misconceptions

With enough time to prepare, the switch cost disappears.

Preparation reduces the cost, but in the classic alternating-runs experiments a residual cost remained even with 1.2 s of preparation, limited to the first trial of the new task (Rogers & Monsell, 1995). The belief exists because the reduction itself is easy to see.

Failing a card-sorting test shows that the frontal lobes are damaged.

The Wisconsin Card Sorting Test, in its traditional form, separates frontal from non-frontal damage poorly, and the brain activity it evokes is widespread (Nyhus & Barceló, 2009). The association lingers because early lesion studies of card sorting focused on the role of the frontal lobes.

Cognitive flexibility is a single, unitary ability.

In 119 young adults, shifting tasks formed five mostly correlated factors, and only some of them could be equated with a general shifting factor (von Bastian & Druey, 2017). The single label persists because flexibility is often measured with one task.

Flexibility is fully developed by early adolescence.

Switching showed protracted development and was still immature at 13 years compared with adults (Davidson et al., 2006). Simple card sorts are passed early in childhood, which makes the capacity look mature sooner than it is.

Glossary

Alternating-runs paradigm.

A switching procedure in which the task changes at a fixed, predictable point, such as every second trial, so that switch and repeat trials can be compared within one block.

Backward inhibition.

The finding that a recently abandoned task is suppressed, so returning to it soon afterwards is harder; it can account for at least part of the residual switch cost.

Cognitive flexibility.

The capacity to change task, rule or strategy when circumstances call for it, usually counted among the executive functions and defined inconsistently across studies.

Cognitive inflexibility.

Difficulty in giving up a rule, task or way of thinking after it has stopped serving; the MeSH entry term for the reverse side of cognitive flexibility.

Dimensional Change Card Sort.

A children's test in which cards are sorted by one dimension and then by another; continuing with the first dimension marks inflexibility.

Extradimensional shift.

A change in which stimulus dimension, rather than which value within a dimension, signals the correct choice.

Mixing cost.

The slowing on repeat trials in blocks that mix two tasks, relative to blocks with one task; also called the global cost.

Perseveration.

Continuing to respond by a rule or dimension after it has ceased to be correct.

Preparation effect.

The reduction in switch cost when more time is available between the cue or previous response and the next stimulus.

Residual switch cost.

The part of the switch cost that remains even after long preparation.

Set shifting.

Changing the rule, feature or mental set that guides responding, typically assessed with card-sorting tasks.

Switch cost.

The slower or less accurate performance on a trial where the task changes, relative to a trial where it repeats; also called the local cost.

Task cuing.

A switching procedure in which a signal before each stimulus tells the participant which task to perform, so the timing of preparation can be controlled.

Task set.

The configuration of rules, relevant features and response mappings needed to perform a particular task.

Trail Making Test.

A paper-and-pencil test whose second part requires alternating between two sequences; performance on it also draws on working memory.

Voluntary task switching.

A procedure in which participants choose which task to perform on each trial, which still yields a switch cost.

Wisconsin Card Sorting Test.

A card-sorting test in which the sorting rule is learned from feedback and changes without warning; in its traditional form, it separates frontal from non-frontal damage poorly.

Key Researchers

Adele Diamond

(University of British Columbia). Developmental cognitive neuroscientist who studies how executive functions, including flexibility, develop and can be trained in childhood. Wikipedia · Wikidata · Faculty page

Andrea Kiesel

(University of Freiburg). Cognitive psychologist who studies how control and interference combine in task switching and multitasking. ORCID · Faculty page

Iring Koch

(RWTH Aachen University). Cognitive psychologist who studies task switching and dual-task performance, including how the two research lines fit together. ORCID · Faculty page

Brenda Milner

(Montreal Neurological Institute, McGill University; b. 1918). Neuropsychologist whose card-sorting study linked perseverative errors to damage of the frontal lobes. Wikipedia · Wikidata · Faculty page

Akira Miyake

(University of Colorado Boulder). Cognitive psychologist who, with colleagues, showed that shifting, updating and inhibition are related but separable executive functions. Faculty page

Stephen Monsell

(University of Exeter). Cognitive psychologist who, with Rogers, introduced the alternating-runs paradigm and showed that preparation shrinks but does not remove the switch cost. ORCID · Faculty page

Lucina Q. Uddin

(University of California, Los Angeles). Cognitive neuroscientist who studies the large-scale brain networks behind cognitive and behavioural flexibility. ORCID · Wikipedia · Wikidata · Faculty page

Frequently Asked Questions

What is cognitive flexibility?

Cognitive flexibility is the capacity to change task, rule or strategy when demands change, and Diamond counts it among three core executive functions along with working memory and inhibition (Diamond, 2013).

What is a switch cost?

A switch cost is the slowing on a trial where the task changes compared with a trial where it repeats; in the classic experiments it shrank with preparation, yet some of it remained even after 1.2 s (Rogers & Monsell, 1995).

Why does switching tasks take extra time?

One major review attributes the cost partly to reconfiguring the system for the new task and partly to interference control, with interference explaining only part of it (Vandierendonck et al., 2010); another judged it not yet clear whether executive control processes are needed to explain switch costs (Kiesel et al., 2010).

Does the Wisconsin Card Sorting Test measure frontal lobe function?

A critical update concluded that the test, in its traditional form, separates frontal from non-frontal damage poorly and that the brain activity it evokes is widespread, so an anatomically pure version of it cannot be built (Nyhus & Barceló, 2009).

Is flexibility separate from other executive functions?

In a latent-variable study of one hundred thirty-seven college students, shifting was moderately correlated with updating and inhibition yet separable from them, although the sample was limited to college students (Miyake et al., 2000).

When does cognitive flexibility mature?

A cross-sectional study of 325 participants found that task switching develops over a long period and was still immature at 13 years compared with adults (Davidson et al., 2006).

Does ageing reduce cognitive flexibility?

A meta-analysis of 36 participant groups found a marked age deficit in the global cost of mixing tasks but no age deficit specific to attention in the local switch cost, which grew multiplicatively with task complexity, using reaction times only (Wasylyshyn et al., 2011).

Can cognitive flexibility be trained?

A review of the evidence describes training results as mixed, with transfer that tends to be narrow and small effects of exercise (Uddin, 2021).

References